US 7483333 B2
A memory device, such as a synchronous random access memory device, includes four banks of memory cells arranged in rows and columns. Different numbers of columns of memory cells are contained in each of the four banks. The bank in which an item of data are stored is determined by either the memory device, a memory controller or a processor based on one or more of several factors. For example, the bank in which the data are stored may be determined by the nature of the data or the length of data bursts written to or read from the memory device. Alternatively, the bank in which the data are stored may be determined based on the source of data written to the memory device or destination for data read from the memory device.
1. A memory device, comprising:
a first bank of memory cells and a second bank of memory cells, each being operable to store data, the first bank containing a number of memory cells that differs from the second bank of memory cells; and
a control logic coupled to the memory cells, the control logic operable to select which bank to write data to and read data from based on a criteria supplied within the memory device itself.
2. The memory device of
3. The memory device of
4. The memory device of
5. The memory device of
6. The memory device of
7. The memory device of
8. The memory device of
9. A memory device, comprising:
a plurality of banks of memory cells, each bank being operable to store data, a first of the banks having a number of memory cells that differs from a number of memory cells in a second of the banks;
a mode register operable to be programmed to select the length of data bursts when the memory device is operating in a burst mode,
a bank control logic coupled to the mode register, the bank control logic operable to select the bank to be accessed based on the length of data burst programmed in the mode register.
10. The memory device of
11. The memory device of
12. The memory device of
13. The memory device of
14. The memory device of
15. The memory device of
16. A memory system, comprising:
a memory device comprising:
a first bank of memory cells and a second bank of memory cells, each being operable to store read and write data, the first bank containing a number of memory cells that differs from the second bank of memory cells; and
a control logic operable to select which bank to write data to or read data from;
a mode register coupled to the control logic, the mode register operable to be programmed to indicate the length of data bursts and the control logic operable to select one of a plurality of banks to write data to and read data from based on the length of the data burst programmed in the mode register; and
a system controller coupled to the memory device, the system controller operable to cause data to be written to and read from the memory device.
17. The memory system of
18. The memory system of
19. The memory system of
20. The memory system of
This application is a continuation of U.S. patent application Ser. No. 11/144,230, filed Jun. 2, 2005, U.S. Pat. No. 7,177,223, which is a divisional of U.S. patent application Ser. No. 10/804,249, filed Mar. 18, 2004, U.S. Pat. No. 7,082,075.
This invention relates to memory devices, and, more particularly, to a memory device that is adapted for optimal performance to carry out each of a variety of functions.
Memory devices, such as dynamic random access memory (“DRAM”) devices, include at least one array or bank of memory cells arranged in rows and columns. A row of memory cells is normally accessed by decoding a row address and activating a corresponding word line that extends through the array or bank. When a word line is activated, all of the memory cells in the corresponding row are accessed, and that row is said to be “open.” The memory cells in one or more columns of the open row are then accessed by decoding a column address and coupling data bits to or from one or more columns corresponding to the decoded column address.
Opening a row of memory cells normally involves coupling each memory cell in the row to one of a respective pair of complementary digit lines that are provided for each column in the array. Coupling a memory cell to a digit line generates a small differential voltage between the digit lines, which are sensed by a sense amplifier provided for the column. The amount of power consumed in opening a row is largely proportional to the number of memory cells in the row since power is consumed in opening each memory cell. As a result of the large number of memory cells that are typically in each row, opening a row of memory cells can consume a significant amount of power. However, expending the power to open an entire row provides the advantage of allowing faster access times since data to be read from the memory cells in the open row are available by simply coupling data bits from respective columns. It is very common to sequentially read data from or write data to the memory cells in all or most of the columns of a row. Thus, the need to open an entire row at a time usually does not result in any wasted power. However, there are other memory operations in which data are read from only a small number of columns in each row. In such cases, opening an entire row, reading data from or writing data to only a few of the memory cells in that row, and then opening a different row needlessly consumes power. For example, if there are 1024 columns in each row, and data bits are read from only 16 of those columns, the amount of power consumed will be approximately 64 times (i.e., 1024/16) the power that would be used to open a row containing only the 16 memory cells that are read.
Not only does opening an entire row to access only a small number of memory cells in the row unnecessarily consume power, but it can also result in slower operation because an array having a large number of columns results in longer word lines having larger capacitances. The larger capacitance of word lines limit the rate at which voltages on the word lines can change to a level that can activate access transistors used to couple respective memory cells to digit lines.
There is therefore a need for a memory device that limits the number of memory cells that are simultaneously opened when only a relatively few memory cells in the open row will be accessed, thereby limiting the needless consumption of power and needless delay in opening rows of memory cells.
A memory device and method in accordance with the invention includes several arrays or banks of memory cells, at least two of which have different numbers of memory cells in each row. Data are preferably stored in the memory arrays according to the type of data being stored. Data of the type that are normally read from sequential addresses are stored in a memory array having a relatively large number of columns. Data of the type that are normally read from random or non-sequential addresses are stored in a memory array having a relatively small number of columns. The array in which the data are stored can be controlled by a variety of means, including the memory device, a system controller coupled to the memory device or a processor coupled to the system controller that is executing an operating system.
A conceptual block diagram of a memory device 10 according to one embodiment of the invention is shown in
The banks 14-28 in which the data are stored is preferably a function of their data type. For example, graphics data, which is normally accessed sequentially over a large range of addresses, are stored in the banks 22, 24 having the largest number of columns. Storing graphics data in the large number of columns contained in these banks 22, 24 results in fast access times since data bits are made available from a large number of columns in an open row, but power tends not to be wasted since the data bits stored in all columns of an open row are accessed. Stack data and heap data, which normally consist of a sequence of addresses, are stored in the banks 14, 16, 18 and 20, respectively, which have relatively few columns. When accessing heap or stack data, addresses are normally not accessed in sequence, so there is no access time penalty in not opening a row containing a large number of columns. But opening a row containing a large number of columns would tend to needlessly consume power because only a relatively few columns would be accessed in each open row. Application data, which are normally accessed in sequences that are longer than the sequence of addresses normally used to access heap and stack data but shorter than the sequence of addresses normally used to access graphics data, are stored in the banks 26, 28 having a moderate number of columns. Storing application data in this manner can provide the optimum balance betweenfast access times and low power consumption.
Although the memory device 10 shown in
A high level block diagram of a dynamic random access memory (“DRAM”) device according to one embodiment of the invention is shown in
Typically, a bank address and a row address is initially received by the address register 104. The bank address is applied to bank control logic 118, and the row address is applied to a row address multiplexer 108. The bank control logic 118 selects the bank to which data are to be written or from which data are to be read by actuating a row address latch and decoder 120 a-d and a column decoder 124 corresponding to a selected bank 130 a-d of memory cells. The bank control logic 118 may select one of four memory banks 130 a-d to which data are to be written or read based on a variety of factors, such as the nature of the data or the source or destination of the data.
The row address multiplexer 108 couples the row address to a number of components associated with the four memory banks 130 a-d depending upon the state of a bank address bit forming part of the row address. Associated with each of the memory banks 130 a-d is respective row address latch/decoder 120 a-d that stores the row address and applies various signals to its respective memory bank 130 a-d as a function of the stored row address. The row address multiplexer 108 also couples row addresses to the row address latches 120 a-d for the purpose of refreshing the memory cells in the memory banks 130 a-d. The row addresses are generated for refresh purposes by a refresh counter 110. As will be described in greater detail below, the row address multiplexer 108 includes address selection circuitry 109 that selects between providing the external address received from the address register 104 and the address received from the refresh counter 110 to the row address latch/decoder 120 a-d as an internal address.
After the row address has been applied to the address register 104 and stored in one of the row address latches/decoder 120 a-d, a column address is applied to the address register 104. The address register 104 couples the column address to a column address latch 112 where the address is stored. Column decoders 124 a-d receive the column addresses from the column address latch 112, and apply various signals to associated column circuitry 126 and respective sense amplifiers 128 a-d for the banks 130 a-d of memory cells. Data to be read from a selected one of the banks 130 a-d are read by actuating a word line in the selected bank, thereby coupling memory cells in the row corresponding to the word line to respective digit lines in the banks 130 a-d, one of which is provided for each column of memory cells in the banks 130 a-d. Voltages stored in the memory cells are detected by the sense amplifiers 128 a-d, one of which is provided for each column of memory cells in each of the banks 130 a-d. Data bits corresponding to the detected voltages are then coupled to the column circuitry 126 and subsequently coupled through an internal read path 140 to a data output register 142, which applies the data to a data bus 150. Data to be written to one of the memory banks 130 a-d are coupled from the data bus 150 through a data input register 154 and through an internal write path 156 to the column circuitry 126, where the data are transferred to memory cells in the memory banks 130 a-d through respective digit lines after actuating a word line for the row of memory cells where the data are to be written.
The SDRAM 100 shown in
With further reference to
The above-described operation of the SDRAM 100 is controlled by control logic 102 responsive to high level command signals received on a control bus. These high level command signals, which are typically generated by a memory controller (not shown), often include a clock enable signal CKE*, a clock signal CLK, a chip select signal CS*, a write enable signal WE*, a row address strobe signal RAS*, and a column address strobe signal CAS*, where the “*”designates the signal as active low. The control logic 102 generates a sequence of command signals responsive to the high level command signals to carry out various memory functions and program memory modes designated by each of the high level command signals, such as memory read, memory write, refresh operations, standby mode, and the like. For example, driving the RAS* and CAS* inputs low with CKE* high will cause the SDRAM 100 to enter a self-refresh mode. In the self-refresh mode, the control logic 102 causes the memory cells corresponding to the address provided by the refresh counter 110 in the banks 130 a-d to be periodically refreshed. These command signals, and the manner in which they accomplish their respective functions, are conventional. Therefore, in the interest of brevity, a further explanation of these control signals will be omitted.
In accordance with another aspect of the invention, the control logic 102 includes a mode register 160 that may be programmed by a user. The user may program the mode register 160, for example, to indicate the length of a burst of data that will be accessed in the SDRAM 100 when the SDRAM 100 operates in a burst mode. In burst-mode SDRAMs, the column address identifies a starting address corresponding to the first of a sequence of columns from which data sequentially will be read or to which data subsequently will be written responsive to each cycle of the clock signal CLK applied to the control logic 102. The burst mode programmed in the mode register 160 by the user can also be used to determine the bank 130 a-d in which the data are stored. If the mode register 160 is programmed to access data with a long burst, the data will be accessed from the BANK3 130 d. If, on the other hand, the mode register is programmed to access data with a short burst, the data will be accessed from either BANK2 130 c or BANK1 130 b. If the SDRAM 100 is not operating in a burst mode, data may be accessed from BANK0 130 a.
Although the SDRAM 100 shown in
The processor 202 is also typically coupled to a cache memory 226, which is usually static random access memory (“SRAM”) and to the SDRAM 100 through the system controller 214. The system controller 214 includes an address bus 234 coupled to the address bus (
According to one embodiment of the invention, the system controller 214 issues a bank address to the SDRAM 100 based on the nature of the data being stored in the SDRAM 100. If the data are the type that will or probably will be accessed sequentially, the system controller can issue a bank address for BANK3 1 30 d. If the data are the type that will or probably will be accessed randomly, the system controller can issue a bank address for BANK0 130 a.
According to another aspect of the invention, thesystem controller 214 issues bank addresses based on the device that is requesting access to the SDRAM 100. For example, if the graphics processor 260 is requesting access to the SDRAM 100, the access data are likely to be stored in a long sequence of addresses. Therefore, the system controller 214 will issue a bank address that will cause the data to be stored in the BANK3 130 d.
According to still another aspect of the invention, the processor 202 executes an operating system that issues bank addresses to access data in the SDRAM 109 depending on the nature of the data being accessed. For example, when the operating system is accessing graphics data, it will issue a bank address that will cause the data to be stored in the BANK3 130 d. On the other hand, when the operating system is accessing heap or stack data, it will issue a bank address that will cause the data to be stored in the BANK0 130 a.
From the foregoing it will be appreciated that, although specific embodiments of the invention have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the invention. Accordingly, the invention is not limited except as by the appended claims.